AI for Legal Teams in 2026: Use Cases, Tools, Ethics Rules and How to Choose

The short answer: legal teams get the most from AI in five places: legal research grounded in real case law, first drafts of documents, contract review against a playbook, document review in litigation and investigations, and monitoring regulatory change. The tools that fit professional work cite their sources, keep client data out of model training, and leave the lawyer in charge of every output. Courts have already sanctioned lawyers who filed AI-invented citations, so verification is not optional. It is part of the job.

This guide is for law firm leaders, general counsel and legal operations teams. It walks through each use case (what the AI does, what it needs, where the lawyer reviews), compares the main legal AI platforms on what their vendors publish, sets out the ethics duties the American Bar Association has described, and ends with a short buying checklist.

Legal AI tools compared

Tool	Main use	What the vendor says it does	Pricing model	Training-data position (vendor’s words)
Harvey	Firm-wide platform for large firms and in-house teams	Agents for end-to-end legal work; Vault for bulk document analysis; Knowledge for legal, regulatory and tax research; Horizon Scanning for regulatory change	Quote only	“We don’t use inputs, outputs, or uploaded documents to train underlying models.”
Legora	Firm-wide platform	Agent, Tabular Review of many documents, Workflows, Legal Research, Word and Outlook add-ins, Monitors for regulation	Quote only	“Zero AI training on your data.”
Thomson Reuters CoCounsel Legal	Research, drafting and review grounded in Westlaw and Practical Law	Deep Research, Westlaw Brief Builder, Tabular Analysis (up to 10,000 documents against 100 questions)	Online configurator; 1, 2 or 3-year terms	“we never use it to train underlying models.”
Lexis+ with Protégé	Research and drafting grounded in LexisNexis content	Drafting, summarising, document analysis in a secure Vault, Shepard’s citation checks, AI workflows	Quote only; Protégé is included in supported LexisNexis products, not sold alone	“We never use customer data to train AI models”
Spellbook	Contract drafting and review in Word	Review, Draft and Associate (an agent for multi-document work)	Quote by number of users; 7-day free trial	“our LLM providers never learn from, train on, or store your data.”
Paxton	Research and drafting for smaller firms and litigators	Drafting, research across US federal and state law, file analysis	Published per-user price (monthly or annual); custom enterprise pricing	Not found on its pricing page
Everlaw	eDiscovery and litigation	Review Assistant (summaries, coding suggestions), Writing Assistant, Deep Dive questions across a document set	Quote only; AI credits for batch and Deep Dive actions	“None of the LLMs will train or fine-tune their models or services on our customers’ data”
Relativity aiR	eDiscovery and investigations	aiR for Review, aiR for Privilege, aiR for Case Strategy, aiR Assist	Pay as you go or 1 and 3-year commitments; AI tools included with committed data volume	Not found

Sources: Harvey security, Legora, CoCounsel Legal, LexisNexis Protégé, Spellbook, Paxton pricing, Everlaw AI framework and Relativity pricing. Relativity has renamed RelativityOne to Relativity aiR, as reported by LawSites in September 2026. For practice management, Clio has built AI into Clio Manage (the successor to Clio Duo); check Clio’s site for which plans include it.

What’s the best AI to use for legal work?

There is no single best tool, because legal work splits into different jobs. A useful way to decide:

• Research that will be cited to a court: use a tool grounded in a legal database with citation checking, such as CoCounsel Legal (Westlaw and Practical Law) or Lexis+ with Protégé (LexisNexis content and Shepard’s).
• Firm-wide drafting, analysis and knowledge work: platforms like Harvey and Legora, which combine agents, bulk document review and firm knowledge.
• Contracts: Word-native tools like Spellbook or contract platforms like Ironclad and Juro. See our AI contract review tools guide.
• Litigation document review: eDiscovery platforms with AI built in, such as Everlaw and Relativity aiR. Browse the eDiscovery software category.

Is there a ChatGPT for legal? And is Claude or ChatGPT better for lawyers?

General assistants such as ChatGPT, Claude, Gemini and Microsoft Copilot are capable at summarising, drafting and reasoning over documents you give them. Their limits for legal work are practical. They are not connected to a citator by default, so they can produce citations that look right and are not. And whether client data is protected depends on the plan and settings your organisation uses. Business and enterprise plans from the major model providers publish their own data-use terms; read them before any client material goes in.

Choosing between general assistants is less important than choosing how you use them. Many legal platforms run on models from several providers and add the legal-specific layer: research databases, citation checking, matter workspaces, ethical walls and audit logs. So the more useful question is which platform your client data lives in and what controls it has, not which chatbot you prefer. For a deeper look at one platform, see our Harvey AI review.

Legal AI use cases, step by step

1. Legal research and memos

• Job: find and summarise the law on a question, with authority.
• Input → AI step → review → output: a research question and jurisdiction → the AI searches a legal database, reasons over the results and drafts an answer with citations → the lawyer reads the cited authorities and checks their treatment → a memo or research note.
• What it needs: a tool grounded in primary law with citation checking; clear jurisdiction and date scope.
• Example tools: CoCounsel Legal Deep Research, Lexis+ with Protégé, Harvey Knowledge, Paxton.
• Measure it: time to first draft memo, and citation error rate found at review.
• Risks: hallucinated or mischaracterised authority. See the benchmark below.

2. Drafting briefs, letters and agreements

• Job: produce a solid first draft fast.
• Input → AI step → review → output: facts, instructions and precedents → the AI drafts in your style and structure → the lawyer rewrites, verifies and signs off → a filing-ready or client-ready document.
• Example tools: Westlaw Brief Builder in CoCounsel Legal, Lexis+ with Protégé drafting, Spellbook Draft, Harvey and Legora agents.
• Measure it: drafting hours per document type, and number of revision rounds.
• Risks: plausible language that misstates facts or law; confidential facts entered into the wrong tool.

3. Contract review and negotiation

• Job: review third-party paper against your playbook and redline it.
• How it works: the AI flags deviations and proposes tracked changes; the lawyer accepts or rejects each. Covered in detail in our contract review guide.
• Example tools: Spellbook, Ironclad Jurist, Juro, LegalOn, and LinkSquares for in-house repositories.

4. Document review in litigation and investigations

• Job: find relevant and privileged documents in large collections and understand the facts.
• Input → AI step → review → output: collected email and files plus review instructions → the AI suggests relevance and privilege coding with reasons, summarises documents and answers questions across the set → review attorneys validate samples and all privilege calls → a production set, privilege log and case chronology.
• What it needs: a clear review protocol, defensible sampling, and agreement with opposing counsel where required.
• Example tools: Relativity aiR for Review and aiR for Privilege; Everlaw Review Assistant and Deep Dive.
• Measure it: documents reviewed per hour, recall and precision on validation samples, and cost per gigabyte reviewed.
• Named example (vendor-reported): in a Relativity press release, Purpose Legal described completing a 300,000-document review in one week with aiR (Relativity, 2025).

5. Regulatory and horizon monitoring

• Job: spot regulatory changes that affect the business or clients.
• Input → AI step → review → output: jurisdictions and topics to watch → the AI scans sources and summarises relevant changes → counsel assesses impact → an alert or client update.
• Example tools: Harvey Horizon Scanning, Legora Monitors.
• Risks: false comfort from silence. A missed alert is not proof nothing changed.

6. In-house intake and matter triage

• Job: route requests from the business to the right lawyer or self-serve answer.
• How it works: requests arrive by form, email or chat → the AI classifies type and urgency and answers routine questions from approved guidance → legal ops reviews routing rules → a triaged queue.
• Example tools: LegalOn lists matter management, intake automation and triage agents on its higher tiers; many teams also use general workflow automation software.

How often do legal AI tools get things wrong?

The best public evidence is a 2024 benchmark by Stanford RegLab and the Stanford Institute for Human-Centered AI. In “AI on Trial” (May 2024), researchers found that Lexis+ AI and Ask Practical Law AI produced incorrect information more than 17% of the time on their test queries, and Westlaw AI-Assisted Research more than 34% of the time. The tools tested were 2024 versions, and vendors have since released new products, so treat the numbers as a snapshot. The lesson still holds: legal-specific tools reduce the problem but do not remove it.

Courts have been clear about who carries the risk:

• Mata v. Avianca (S.D.N.Y., June 2023). Judge P. Kevin Castel sanctioned two lawyers and their firm $5,000 for submitting non-existent opinions with fake quotes and citations generated by ChatGPT, and for standing by them after the court questioned them. The court noted there is “nothing inherently improper about using a reliable artificial intelligence tool for assistance,” but that lawyers have a gatekeeping role to ensure accuracy (sanctions order).
• Johnson v. Dunn (N.D. Ala., July 2025). Judge Anna M. Manasco publicly reprimanded three attorneys whose motions contained false citations from ChatGPT, disqualified them from the case and referred them to the state bar. The court found fines and a reprimand alone would not deter the conduct. Notably, their firm already had AI policies in place (opinion).

What does ABA Formal Opinion 512 require?

The ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512 on generative AI tools on July 29, 2024 (PDF). In plain terms, it maps existing Model Rules to AI use:

Duty	What it means for AI use
Competence (Rule 1.1)	Understand what the tool can and cannot do, and check its output. You need not be an expert, but you cannot rely on it uncritically.
Confidentiality (Rule 1.6)	Assess the risk before entering client information. For self-learning tools that may reuse inputs, the opinion says the client’s informed consent is required first.
Communication (Rule 1.4)	Whether to tell the client depends on the circumstances; if they ask, answer. Engagement letters are a logical place for disclosure.
Candor and meritorious claims (Rules 3.1, 3.3)	You are responsible for every citation and argument you file, whoever or whatever drafted it.
Supervision (Rules 5.1, 5.3)	Firm leaders must set AI policies and train and supervise lawyers, staff and vendors.
Fees (Rule 1.5)	Hourly bills must reflect actual time, so efficiency gains flow to the client, and you should not bill for learning a general-purpose tool.

State bars have issued their own guidance too. Check the rules where you practise.

Can I use AI as my lawyer?

For understanding a document or preparing questions, AI can help anyone. For advice you will rely on, no. AI tools are not licensed to practise law, and conversations with them are not privileged. Paxton’s own pricing page makes this point, noting that communications with it are protected by its privacy policy “but not by the attorney-client privilege or as work product.” If the stakes are real, use AI to prepare, then talk to a lawyer.

How much does legal AI cost?

Most enterprise legal AI is sold by quote. Harvey, Legora, Spellbook and Everlaw publish no prices. LexisNexis says Protégé pricing is customised by organisation size and needs. Thomson Reuters uses an online configurator based on sector, number of attorneys and term length. Relativity prices by data volume, with its AI tools included in committed volume. Paxton is an exception: it publishes a per-user monthly and annual price on its pricing page and offers custom enterprise pricing. When comparing quotes, check whether AI is an add-on or included, whether usage is capped (Everlaw uses AI credits for batch actions), and the contract term.

Enterprise and security checklist

• Training and retention written into the contract, including the vendor’s model providers.
• Certifications as stated by the vendor: Harvey lists SOC 2 Type II, ISO 27001, ISO 27701 and ISO 42001; Legora lists SOC 2 Type II, ISO 27001, ISO 42001 and HIPAA; CoCounsel Legal states SOC 2 and ISO 42001; Spellbook states SOC 2 Type II.
• Access controls: SAML SSO, SCIM, matter-level permissions and ethical walls. Harvey states SAML SSO, audit logs and IP allow-listing.
• Data residency: Harvey offers EU and Switzerland or Australia processing. Ask the others.
• Document management integration: CoCounsel Legal connects to iManage, NetDocuments, SharePoint and HighQ, per Thomson Reuters.

How to choose legal AI

1. Pick two or three use cases with measurable volume, such as NDA review or first-draft research memos.
2. Write an AI policy covering approved tools, client consent, verification and billing, in line with Opinion 512.
3. Pilot with real matters (with consent where needed) and a senior reviewer checking every output.
4. Measure hours saved, error rates found in review and lawyer adoption.
5. Then scale, with training and a named owner for the tool.

For case and practice management around these tools, see our guide to legal case management software and the legal practice management category.

Related guides: for the wider picture, see our guides to AI governance tools, how to implement AI in business and enterprise AI use cases by function.

Frequently asked questions